BMS terminal voltage identification method based on WPD-GMDH neural network

Abstract

The invention provides a BMS terminal voltage identification method based on a WPD-GMDH neural network, and belongs to the technical field of lithium battery management system identification. The technical problems of high difficulty and low precision of identifying the lithium battery management system are solved. According to the technical scheme, the method comprises the following steps of 1) obtaining sample data through a simulation experiment; 2) preprocessing data, constructing a data set, and dividing the data set into a training set and a test set; and step 3) training the data set through the WPD-GMDH neural network to obtain a WPD-GMDH model for real-time estimation. The beneficial effects of the invention are that the WPD-GMDH neural network used in the invention separates signals of different frequencies through the WPD technology, thereby finding the nonlinear relation of different features, and improving the learning capability of the GMDH neural network model.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery management systems, and in particular to a BMS terminal voltage identification method based on a WPD-GMDH neural network. Background Art

[0002] With the development of new energy and lithium battery industries, the requirements for the measurement of the terminal voltage of the battery management system (BMS) are getting higher and higher. BMS not only needs to be able to charge and discharge quickly, but also the accuracy of the BMS terminal voltage is getting higher and higher. In order to better analyze and control the BMS terminal voltage, it is necessary to establish a corresponding mathematical model for the BMS terminal voltage.

[0003] The existing BMS terminal voltage identification methods include particle swarm optimization (PSO), convolutional neural network (CNN) and long short-term memory neural network (LSTM). For example, in the paper "Identification of main steam temperature system based on improved particle swarm algorithm", although the improved PSO algorithm was successfully used to obtain the optimal model parameters of the main steam temperature system, the industrial system was disturbed by various factors such as loss and model mismatch during actual operation, and parameter identification was impossible. In the paper "Identification of power system operation status based on particle swarm optimization and convolutional neural network", CNN successfully identified the real-time operation status of the power system with high precision, but the system scalability is still insufficient, data feature extraction needs further refinement and research, and there is also a problem of high computational complexity. In the article "Comparative Study of Deep Recurrent Neural Networks in Ship Maneuvering Motion Identification", although LSTM and GRU networks were successfully applied to ship maneuvering motion identification modeling, when faced with more complex systems or longer measured data, the identification accuracy may decrease.

[0004] BMS is a complex and strongly nonlinear system, some parts of which cannot be expressed by mathematical expressions, and the operating conditions of BMS are complex and changeable, so it is impossible to obtain its accurate model. Therefore, the first three methods are not comprehensive, and neural network system identification methods such as GMDH are more suitable for BMS terminal voltage identification. GMDH neural network has developed rapidly due to its high precision and strong adaptability. By introducing wavelet packet decomposition (WPD) technology in GMDH, highlighting the influence of key features and mining the complex nonlinear relationship between input and output, it can not only overcome the long-term dependence problem, but also solve the technical problems of difficulty and low precision in identifying BMS. However, since GMDH is a technology based on the principle of multi-layer neural network and variety improvement hypothesis, it has become a difficulty to use it for BMS terminal voltage identification.

[0005] How to solve the above technical problems is the subject faced by the present invention. Summary of the invention

[0006] The purpose of the present invention is to provide a BMS terminal voltage identification method based on the WPD-GMDH neural network. The input and output data are obtained through the BMS model simulation experiment, and then the data is processed and a training data set is constructed. The training set is imported into the WPD-GMDH neural network for training, which can finally be used for the identification of the BMS terminal voltage.

[0007] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is specifically: a BMS terminal voltage identification method based on WPD-GMDH neural network, which specifically includes the following steps:

[0008] Step 1) obtaining sample data through simulation experiments;

[0009] Step 2) preprocess the data and construct the data set, and divide it into a training set and a test set;

[0010] Step 3) training the WPD-GMDH neural network using the training set to obtain a WPD-GMDH model for online identification;

[0011] As a further optimization scheme of the BMS terminal voltage identification method based on the WPD-GMDH neural network provided by the present invention, the step 2) specifically includes the following steps:

[0012] Step 2-1) When using the neural network method for identification, the measured data needs to be normalized to avoid problems such as slow convergence speed and high error caused by the lack of unified dimensions of the data. Min-Max standardization is used to normalize the input data;

[0013] Step 2-2) Further, the normalized data is divided into a training set and a test set;

[0014] As a further optimization scheme of the BMS terminal voltage identification method based on the WPD-GMDH neural network provided by the present invention, the step 3) specifically includes the following steps:

[0015] Step 3-1) GMDH neural network is a method for identifying complex nonlinear systems based on the principle of multi-layer neural network and the hypothesis of variety improvement, and using Kolmogorov-Gavbor polynomials as the basis for self-organizing modeling. In order to improve the performance of GMDH and the learning ability of the model, wavelet packet decomposition (WPD) technology is introduced into the GMDH model.

[0016] The specific optimization methods are as follows:

[0017] WPD is an emerging technology that decomposes nonlinear signals through high resolution. After the data is decomposed by WPD, modal functions with different frequency characteristics are obtained. WPD can quickly decompose the modal function into high-frequency, low-frequency and residual components, and combine them into a feature matrix as the input of GMDH. The nonlinear relationship in the data is mined and identified in GMDH, and finally the identification results are superimposed to obtain a complete BMS terminal voltage identification result.

[0018] The data components processed by WPD are identified using the GMDH model, and multiple modal data are input into the GMDH model as feature information for training and identification. Finally, the results of the identified multiple components are superimposed to obtain the final identification result of the BMS terminal voltage.

[0019] After the WPD-GMDH identification model is trained, the test sample is substituted into the WPD-GMDH neural network model, and then the output sequence in the test sample is substituted into the trained WPD-GMDH model. Finally, the model operation results and the output sequence in the test sample are compared. The root mean square error (RMSE) is used as the evaluation method to compare the model operation results and the output sequence in the test sample.

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] (1) The WPD-GMDH identification model of the present invention includes a WPD decomposition layer, an input layer, a fully connected layer and an identification value. The WPD technology is introduced into the GMDH model, which can effectively highlight different nonlinear relationships, thereby improving the identification effect, and can effectively improve the generalization ability and accuracy of the GMDH model. Finally, the WPD-GMDH identification model is obtained. While decomposing the low-frequency subset of the signal, the WPD continues to decompose the high-frequency subset, and can set the number of decomposition layers and the wavelet function used for decomposition according to the decomposition needs.

[0022] (2) Considering that the BMS is disturbed by various factors such as loss during actual operation, there is no fixed specific mathematical model. The WPD-GMDH model has the advantages of high accuracy and strong fitting ability, and does not require multiple trainings, which can save a lot of time. The method proposed in the present invention can make a more accurate identification of the BMS terminal voltage, which has certain theoretical research and engineering practical significance.

[0023] (3) The GMDH network is different from other neural networks in that the GMDH network structure is not fixed and keeps changing during the training process. This feature is also one of the main reasons for choosing the GMDH neural network for BMS terminal voltage identification. By continuously screening and combining to establish a model of the nonlinear system, it is very effective in identifying high-order nonlinear systems and can effectively improve the BMS terminal voltage identification accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0025] Figure 1 This is a physical picture of the BMS in the present invention.

[0026] Figure 2 This is a flow chart of a BMS terminal voltage identification method based on a WPD-GMDH neural network in the present invention.

[0027] Figure 3 This is the structural diagram of the WPD-GMDH identification model in the present invention.

[0028] Figure 4 This is a diagram showing the result of WPD-GMDH identifying the BMS terminal voltage in the present invention. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. Of course, the specific embodiments described here are only used to explain the present invention and are not used to limit the present invention.

[0030] Example

[0031] See also Figures 1 to 4 The technical solution provided by this embodiment is: a BMS terminal voltage identification method based on WPD-GMDH neural network. This embodiment conducts research on the BMS terminal voltage.

[0032] The proposed BMS terminal voltage identification method based on WPD-GMDH neural network is applied to a lithium battery management system. The composition distribution diagram of the system is shown in Figure 1 .

[0033] MATLAB2021b software is used in the identification process. In order to better achieve the purpose of this embodiment, this embodiment is a BMS terminal voltage identification method based on the WPD-GMDH neural network, including the following steps:

[0034] Step 1) obtaining sample data through simulation experiments;

[0035] Step 2) preprocess the data and construct the data set, and divide it into a training set and a test set;

[0036] Step 3) The WPD-GMDH neural network is trained with the training set to obtain a WPD-GMDH model for online identification.

[0037] Specifically, step 2) includes the following steps:

[0038] Step 2-1) When using the neural network method for identification, the measured data needs to be normalized to avoid problems such as slow convergence and high error caused by the lack of uniform dimension of the data. Min-Max standardization is used to normalize the input data.

[0039] Step 2-2) Further, the normalized data is divided into a training set and a test set.

[0040] Specifically, step 3) comprises the following steps:

[0041] Step 3-1) GMDH neural network is a method for identifying complex nonlinear systems based on the principle of multi-layer neural network and the hypothesis of variety improvement, and using Kolmogorov-Gavbor polynomials as the basis for self-organizing modeling. In order to improve the performance of GMDH and the learning ability of the model, wavelet packet decomposition (WPD) technology is introduced into the GMDH model.

[0042] The data components processed by WPD are identified using the GMDH model, and multiple modal data are input into the GMDH model as feature information for training and identification. Finally, the results of the identified multiple components are superimposed to obtain the final identification result of the BMS terminal voltage.

[0043] After the WPD-GMDH identification model is trained, the test sample is substituted into the WPD-GMDH neural network model, and then the output sequence in the test sample is substituted into the trained WPD-GMDH model. Finally, the model operation results and the output sequence in the test sample are compared. The root mean square error (RMSE) is used as the evaluation method to compare the model operation results and the output sequence in the test sample.

[0044] Finally, the BMS terminal voltage identification result is as follows: Figure 4 The model evaluation index root mean square error (RMSE) is 0.0084, and the actual output curve value is close to the output curve value of the WPD-GMDH model, indicating that the effect of the model is stable and the identification accuracy is high.

[0045] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A BMS terminal voltage identification method based on WPD-GMDH neural network, characterized in that: The following steps are involved: Step 1) obtaining sample data through simulation experiments; Step 2) preprocess the data and construct the data set, and divide it into a training set and a test set; Step 3) The WPD-GMDH neural network is trained with the training set to obtain a WPD-GMDH model for online identification.

2. According to claim 1, a BMS terminal voltage identification method based on WPD-GMDH neural network is characterized in that: The step 2) comprises the following steps: Step 2-1) During the neural network method identification, the measured data is normalized, and the input data is normalized using Min-Max standardization; Step 2-2) Divide the normalized data into training set and test set.

3. The BMS terminal voltage identification method based on WPD-GMDH neural network according to claim 1 is characterized in that: The step 3) comprises the following steps: Step 3-1) The GMDH neural network is based on the principle of multi-layer neural network and the variety improvement hypothesis, and uses the Kolmogorov-Gavbor polynomial as the basis for self-organizing modeling.

4. The BMS terminal voltage identification method based on WPD-GMDH neural network according to claim 1 is characterized in that: In step 3-1), the specific optimization method includes the following steps: Step 3-1-1) After the data is decomposed by WPD, modal functions with different frequency characteristics are obtained. WPD decomposes the modal function into high-frequency, low-frequency and residual components, and combines them into a feature matrix as the input of GMDH. The matrix is ​​input into GMDH to mine the nonlinear relationship in the data for identification. Finally, the identification results are superimposed to obtain the complete BMS terminal voltage identification result.

5. The BMS terminal voltage identification method based on WPD-GMDH neural network according to claim 1 is characterized in that: In the step 3-1-1), the signal is multi-scaled refined by scaling and translation operations. WPD decomposes the low-frequency subset of the signal while continuing to decompose the high-frequency subset. The number of decomposition layers and the wavelet function used for decomposition are set according to the decomposition needs. It has been widely used in various time series data decompositions.

6. The BMS terminal voltage identification method based on WPD-GMDH neural network according to claim 1 is characterized in that: In the step 3-1-1), the data components processed by WPD are identified using the GMDH model, and multiple modal data are input into the GMDH model as feature information for training and identification. Finally, the results of the identified multiple components are superimposed to obtain the final identification result of the BMS terminal voltage.

7. The BMS terminal voltage identification method based on WPD-GMDH neural network according to claim 1 is characterized in that: In the step 3-1-1), after the WPD-GMDH recognition model is trained, the test sample is substituted into the WPD-GMDH neural network model, and then the output sequence in the test sample is substituted into the trained WPD-GMDH model. Finally, the model operation results and the output sequence in the test sample are compared, and the root mean square error RMSE is used as an evaluation method to compare the model operation results and the output sequence in the test sample.

Images